14360: Fix tests.
[arvados.git] / services / crunch-run / crunchrun.go
index 098c53f8a6a587816703ad6997ceb51eec7f0232..27fb8367f50e05764d45c77dc362497f7f702847 100644 (file)
@@ -91,37 +91,39 @@ type PsProcess interface {
 // ContainerRunner is the main stateful struct used for a single execution of a
 // container.
 type ContainerRunner struct {
-       Docker    ThinDockerClient
-       client    *arvados.Client
-       ArvClient IArvadosClient
-       Kc        IKeepClient
-       arvados.Container
+       Docker          ThinDockerClient
+       client          *arvados.Client
+       ArvClient       IArvadosClient
+       Kc              IKeepClient
+       Container       arvados.Container
        ContainerConfig dockercontainer.Config
-       dockercontainer.HostConfig
-       token       string
-       ContainerID string
-       ExitCode    *int
-       NewLogWriter
-       loggingDone   chan bool
-       CrunchLog     *ThrottledLogger
-       Stdout        io.WriteCloser
-       Stderr        io.WriteCloser
-       LogCollection arvados.CollectionFileSystem
-       LogsPDH       *string
-       RunArvMount
-       MkTempDir
-       ArvMount      *exec.Cmd
-       ArvMountPoint string
-       HostOutputDir string
-       Binds         []string
-       Volumes       map[string]struct{}
-       OutputPDH     *string
-       SigChan       chan os.Signal
-       ArvMountExit  chan error
-       SecretMounts  map[string]arvados.Mount
-       MkArvClient   func(token string) (IArvadosClient, error)
-       finalState    string
-       parentTemp    string
+       HostConfig      dockercontainer.HostConfig
+       token           string
+       ContainerID     string
+       ExitCode        *int
+       NewLogWriter    NewLogWriter
+       loggingDone     chan bool
+       CrunchLog       *ThrottledLogger
+       Stdout          io.WriteCloser
+       Stderr          io.WriteCloser
+       logUUID         string
+       logMtx          sync.Mutex
+       LogCollection   arvados.CollectionFileSystem
+       LogsPDH         *string
+       RunArvMount     RunArvMount
+       MkTempDir       MkTempDir
+       ArvMount        *exec.Cmd
+       ArvMountPoint   string
+       HostOutputDir   string
+       Binds           []string
+       Volumes         map[string]struct{}
+       OutputPDH       *string
+       SigChan         chan os.Signal
+       ArvMountExit    chan error
+       SecretMounts    map[string]arvados.Mount
+       MkArvClient     func(token string) (IArvadosClient, error)
+       finalState      string
+       parentTemp      string
 
        ListProcesses func() ([]PsProcess, error)
 
@@ -742,6 +744,7 @@ func (runner *ContainerRunner) startCrunchstat() error {
                CgroupParent: runner.expectCgroupParent,
                CgroupRoot:   runner.cgroupRoot,
                PollPeriod:   runner.statInterval,
+               TempDir:      runner.parentTemp,
        }
        runner.statReporter.Start()
        return nil
@@ -1174,6 +1177,54 @@ func (runner *ContainerRunner) WaitFinish() error {
        }
 }
 
+func (runner *ContainerRunner) updateLogs() {
+       ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
+       defer ticker.Stop()
+
+       sigusr1 := make(chan os.Signal, 1)
+       signal.Notify(sigusr1, syscall.SIGUSR1)
+       defer signal.Stop(sigusr1)
+
+       saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
+       saveAtSize := crunchLogUpdateSize
+       var savedSize int64
+       for {
+               select {
+               case <-ticker.C:
+               case <-sigusr1:
+                       saveAtTime = time.Now()
+               }
+               runner.logMtx.Lock()
+               done := runner.LogsPDH != nil
+               runner.logMtx.Unlock()
+               if done {
+                       return
+               }
+               size := runner.LogCollection.Size()
+               if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
+                       continue
+               }
+               saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
+               saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
+               saved, err := runner.saveLogCollection(false)
+               if err != nil {
+                       runner.CrunchLog.Printf("error updating log collection: %s", err)
+                       continue
+               }
+
+               var updated arvados.Container
+               err = runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
+                       "container": arvadosclient.Dict{"log": saved.PortableDataHash},
+               }, &updated)
+               if err != nil {
+                       runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
+                       continue
+               }
+
+               savedSize = size
+       }
+}
+
 // CaptureOutput saves data from the container's output directory if
 // needed, and updates the container output accordingly.
 func (runner *ContainerRunner) CaptureOutput() error {
@@ -1311,26 +1362,51 @@ func (runner *ContainerRunner) CommitLogs() error {
                // -- it exists only to send logs to other channels.
                return nil
        }
+       saved, err := runner.saveLogCollection(true)
+       if err != nil {
+               return fmt.Errorf("error saving log collection: %s", err)
+       }
+       runner.logMtx.Lock()
+       defer runner.logMtx.Unlock()
+       runner.LogsPDH = &saved.PortableDataHash
+       return nil
+}
 
+func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
+       runner.logMtx.Lock()
+       defer runner.logMtx.Unlock()
+       if runner.LogsPDH != nil {
+               // Already finalized.
+               return
+       }
        mt, err := runner.LogCollection.MarshalManifest(".")
        if err != nil {
-               return fmt.Errorf("While creating log manifest: %v", err)
-       }
-
-       var response arvados.Collection
-       err = runner.ArvClient.Create("collections",
-               arvadosclient.Dict{
-                       "ensure_unique_name": true,
-                       "collection": arvadosclient.Dict{
-                               "is_trashed":    true,
-                               "name":          "logs for " + runner.Container.UUID,
-                               "manifest_text": mt}},
-               &response)
+               err = fmt.Errorf("error creating log manifest: %v", err)
+               return
+       }
+       updates := arvadosclient.Dict{
+               "name":          "logs for " + runner.Container.UUID,
+               "manifest_text": mt,
+       }
+       if final {
+               updates["is_trashed"] = true
+       } else {
+               exp := time.Now().Add(crunchLogUpdatePeriod * 24)
+               updates["trash_at"] = exp
+               updates["delete_at"] = exp
+       }
+       reqBody := arvadosclient.Dict{"collection": updates}
+       if runner.logUUID == "" {
+               reqBody["ensure_unique_name"] = true
+               err = runner.ArvClient.Create("collections", reqBody, &response)
+       } else {
+               err = runner.ArvClient.Update("collections", runner.logUUID, reqBody, &response)
+       }
        if err != nil {
-               return fmt.Errorf("While creating log collection: %v", err)
+               return
        }
-       runner.LogsPDH = &response.PortableDataHash
-       return nil
+       runner.logUUID = response.UUID
+       return
 }
 
 // UpdateContainerRunning updates the container state to "Running"
@@ -1356,7 +1432,7 @@ func (runner *ContainerRunner) ContainerToken() (string, error) {
        if err != nil {
                return "", err
        }
-       runner.token = auth.APIToken
+       runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
        return runner.token, nil
 }
 
@@ -1427,11 +1503,11 @@ func (runner *ContainerRunner) Run() (err error) {
                // hasn't already been assigned when Run() returns,
                // this cleanup func will cause Run() to return the
                // first non-nil error that is passed to checkErr().
-               checkErr := func(e error) {
+               checkErr := func(errorIn string, e error) {
                        if e == nil {
                                return
                        }
-                       runner.CrunchLog.Print(e)
+                       runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
                        if err == nil {
                                err = e
                        }
@@ -1442,7 +1518,7 @@ func (runner *ContainerRunner) Run() (err error) {
                }
 
                // Log the error encountered in Run(), if any
-               checkErr(err)
+               checkErr("Run", err)
 
                if runner.finalState == "Queued" {
                        runner.UpdateContainerFinal()
@@ -1455,10 +1531,10 @@ func (runner *ContainerRunner) Run() (err error) {
                        // capture partial output and write logs
                }
 
-               checkErr(runner.CaptureOutput())
-               checkErr(runner.stopHoststat())
-               checkErr(runner.CommitLogs())
-               checkErr(runner.UpdateContainerFinal())
+               checkErr("CaptureOutput", runner.CaptureOutput())
+               checkErr("stopHoststat", runner.stopHoststat())
+               checkErr("CommitLogs", runner.CommitLogs())
+               checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
        }()
 
        err = runner.fetchContainerRecord()
@@ -1629,6 +1705,7 @@ func NewContainerRunner(client *arvados.Client, api IArvadosClient, kc IKeepClie
        cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
 
        loadLogThrottleParams(api)
+       go cr.updateLogs()
 
        return cr, nil
 }
@@ -1639,6 +1716,10 @@ func main() {
        cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
        cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
        caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
+       detach := flag.Bool("detach", false, "Detach from parent process and run in the background")
+       sleep := flag.Duration("sleep", 0, "Delay before starting (testing use only)")
+       kill := flag.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
+       list := flag.Bool("list", false, "List UUIDs of existing crunch-run processes")
        enableNetwork := flag.String("container-enable-networking", "default",
                `Specify if networking should be enabled for container.  One of 'default', 'always':
        default: only enable networking if container requests it.
@@ -1650,8 +1731,29 @@ func main() {
        memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
        getVersion := flag.Bool("version", false, "Print version information and exit.")
        checkContainerd := flag.Duration("check-containerd", 60*time.Second, "Periodic check if (docker-)containerd is running (use 0s to disable).")
+
+       detached := false
+       if len(os.Args) > 1 && os.Args[1] == "-detached" {
+               // This process was invoked by a parent process, which
+               // has passed along its own arguments, including
+               // -detach, after the leading -detached flag.  Strip
+               // the leading -detached flag (it's not recognized by
+               // flag.Parse()) ... and remember not to detach all
+               // over again in this process.
+               os.Args = append([]string{os.Args[0]}, os.Args[2:]...)
+               detached = true
+       }
        flag.Parse()
 
+       switch {
+       case *detach && !detached:
+               os.Exit(Detach(flag.Arg(0), os.Args, os.Stdout, os.Stderr))
+       case *kill >= 0:
+               os.Exit(KillProcess(flag.Arg(0), syscall.Signal(*kill), os.Stdout, os.Stderr))
+       case *list:
+               os.Exit(ListProcesses(os.Stdout, os.Stderr))
+       }
+
        // Print version information if requested
        if *getVersion {
                fmt.Printf("crunch-run %s\n", version)
@@ -1659,6 +1761,7 @@ func main() {
        }
 
        log.Printf("crunch-run %s started", version)
+       time.Sleep(*sleep)
 
        containerId := flag.Arg(0)